Skip to content

fix(rollout): skip process cleanup for Oracle - #1085

Open
zhiheng-yang wants to merge 1 commit into
benchflow-ai:mainfrom
zhiheng-yang:fix/skip-oracle-process-cleanup
Open

fix(rollout): skip process cleanup for Oracle#1085
zhiheng-yang wants to merge 1 commit into
benchflow-ai:mainfrom
zhiheng-yang:fix/skip-oracle-process-cleanup

Conversation

@zhiheng-yang

@zhiheng-yang zhiheng-yang commented Aug 31, 2026

Copy link
Copy Markdown

fix(rollout): skip process cleanup for Oracle

Summary

Oracle mode awaits solve.sh directly and does not start an ACP or session agent. Rollout.disconnect() nevertheless treated the fallback launch label oracle as a process name and derived this cleanup command:

pkill -f '(^|[ /])oracle( |$)'

There is no persistent Oracle process for that pattern to clean up. On a runtime that shares the runner's PID namespace, it can instead match BenchFlow's own --agent oracle command and terminate the run after verification but before the final result is built.

This change skips pattern-based cleanup only when the current launch value is exactly oracle. Other agents, including similarly named launches such as oracle-agent-acp, keep the existing cleanup behavior. The check uses the current _agent_launch, so role changes through connect_as() remain aligned with the process that was actually launched.

Testing

  • Added regression coverage for oracle, opencode, and the oracle-agent-acp near-match; the regression fails on main and passes with this change.
  • Ran the relevant rollout and lifecycle tests: 286 passed. Ruff, changed-file formatting, and ty check also pass.
  • Completed an end-to-end SkillsBench Oracle run on Daytona as a non-regression check: 1/1 passed, mean reward 1.00, errors=0, with the final summary and artifacts written successfully.

Devin Review

@devin-ai-integration devin-ai-integration Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.

Devin Review

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant